min(axis = 1) propagates NAN when encountering missing timestamps

[randomgambit]

Hello there,
Consider this:

dftest = pd.DataFrame({'time1' : ['2019-01-01- 10:10:10.123456Z',
                                  '2019-01-01- 10:10:10.123456Z'],
                       'time2' : ['ha',
                                 'pandas!!!'],
                       'time3' :['here we go',
                                'again']})

dftest
Out[179]: 
                          time1      time2       time3
0  2019-01-01- 10:10:10.123456Z         ha  here we go
1  2019-01-01- 10:10:10.123456Z  pandas!!!       again

I am trying to get the minimum timestamp between time1, time2 and time3. Of course, in this example time2 and time3 are not timestamps so (after proper conversion to NaTs) I expect the minimum to be time1 all the time.

Surprisingly, this fails entirely.

dftest['time1'] = pd.to_datetime(dftest.time1, errors = 'coerce')
dftest['time2'] = pd.to_datetime(dftest.time2, errors = 'coerce')    
dftest['time3'] = pd.to_datetime(dftest.time3, errors = 'coerce') 
dftest
Out[181]: 
                             time1 time2 time3
0 2019-01-01 10:10:10.123456+00:00   NaT   NaT
1 2019-01-01 10:10:10.123456+00:00   NaT   NaT

And now:

dftest.loc[:, ['time1', 'time2', 'time3']].min(axis = 1)
Out[182]: 
0   nan
1   nan
dtype: float64

This looks like a bug. It seems that the errors is coming from the timezone component. What do you think?

Thanks!

[mroeschke]

Thanks for the report. This looks like a duplicate of #27794